Oil Pipeline Work Conditions Clustering Based on Simulated Annealing K-Means Algorithm

Abstract

With regards to the characteristics of work conditions on oil pipeline, such as complicated changes, lack of prior knowledge and difficult classification, simulated annealing K-means clustering algorithm are proposed. Samples, which include various work condition changes of oil pipeline, are selected from pressure data collected in field. In order to analyze data conveniently, each group of raw data is normalized with mean zero and de-noised with wavelet transform. Eigenvectors can be used in clustering analysis; they are composed of time-domain statistical indexes. Clustering centers can be attained by iterative computation with K-means algorithm. The principle of K-means algorithm is that square sum, between all samples in cluster domain and cluster centers, is minimum. To fulfill K-means algorithm is simple and the convergence is fast; meanwhile, it has some limitations. Simulated annealing algorithm is based on randomized searching algorithm and global optimization algorithm. By employing the optimize algorithm, the local minimum question of K-means algorithm can be avoided. The cluster result of K-means algorithm is used as initial solution; as a result, the optimal cluster centers are attained by simulated annealing. In the field, it has been well verified that the optimal cluster centers as evaluation standard of pipeline operation conditions.